set temperature 0.3

app.py

@@ -19,7 +19,7 @@ model = LlavaForConditionalGeneration.from_pretrained(
     model_id,
     quantization_config=quantization_config,
     device_map="auto",
-    use_flash_attention_2=True,
+    # use_flash_attention_2=True,
     low_cpu_mem_usage=True
 )
 
@@ -28,7 +28,7 @@ def text_to_image(image, prompt):
     prompt = f'USER: <image>\n{prompt}\nASSISTANT:'
 
     inputs = processor([prompt], images=[image], padding=True, return_tensors="pt").to(model.device)
-    output = model.generate(**inputs, max_new_tokens=500)
+    output = model.generate(**inputs, max_new_tokens=500, temperature=0.3)
     generated_text = processor.batch_decode(output, skip_special_tokens=True)
     text = generated_text.pop()
     text_output = text.split("ASSISTANT:")[-1]